Revisiting Regional Trading Agreements with Proper ...publi.cerdi.org/ed/2002/2002.10.pdf ·...
Transcript of Revisiting Regional Trading Agreements with Proper ...publi.cerdi.org/ed/2002/2002.10.pdf ·...
CENTRE D’ETUDES ET DE RECHERCHES SUR LE DEVELOPPEMENT INTERNATIONAL
Document de travail de la série
Etudes et Documents E 2002.10
Revisiting Regional Trading Agreements
with Proper Specification of the Gravity Model
Céline CARRERE*
May 2002, 33 p.
* CERDI : Centre d’études et de recherches sur le développement international,
65 boulevard François Mitterrand, 63 000 Clermont Ferrand – France.
Tel : (33) 473 177 400. Fax : (33) 473 177 428. E.mail : [email protected]. I thank J.F Brun, Patrick Guillaumont and Jaime de Melo for guidance and comments, as well as Alan Winters for valuable comments.
1
Abstract
This paper uses a gravity model to assess ex-post regional trade agreements. The model includes 130 countries and is estimated in panel over the period 1962-96. The introduction of the correct number of dummy variables allows for identification of Vinerian trade creation and trade diversion effects, while the estimation method takes into account a potential correlation between the explanatory variables and the bilateral specific effects introduced in the model, as well as potential selection bias. In contrast with previous estimates, results show that over the period 1962-1996, regional agreements have generated a significant increase in trade between members, often at the expense of the rest of the world. JEL Classification: F11; F15; C23. Keywords: Regional trade agreements, Gravity equation, Trade creation, Trade diversion, Panel Data.
Résumé
Ce papier utilise un modèle de gravité pour évaluer ex-post des accords commerciaux régionaux. Le modèle est estimé en panel, sur 130 pays et sur la période 1962-96. L’introduction du nombre correct de variables muettes permet d’identifier les effets de création et de détournement de trafic vinérien, selon une méthode d’estimation qui prend en compte (i) la corrélation potentielle entre certaines variables explicatives et les effets spécifiques bilatéraux introduits dans le modèle, (ii) un biais de sélection potentiel. Contrairement aux estimations des études précédentes, les résultats mettent en évidence que sur la période 1962-1996, les accords régionaux considérés dans ce papier ont engendré une augmentation significative du commerce entre les pays membres, souvent au détriment du reste du monde.
JEL Classification: F11; F15; C23. Mots-clé : Accords Commerciaux Régionaux, Equation de gravité, création de trafic, détournement de trafic, données de panel.
2
1. Introduction
After a long period of neglect following its introduction in the late sixties (Poyhonen,
1963, Tinbergen, 1962, Linnemann, 1966, Aitken, 1973) since the late eighties, the
gravity trade model has acquired a second youth. First, it discovered new theoretical
foundations both with the advent of the trade theories based on monopolistic
competition and firm-level product differentiation which predict that the intensity of
trade should be inversely related to GDP across trading partners (Krugman and
Helpman, 1985, Feenstra, Markusen and Rose, 1998) and within a perfect competition
setting with product differentiation at the national level (Deardorff, 1998). Second, the
gravity model is used extensively to study trade patterns, as for example in the case of
the drastic changes following the demise of central planning. Most recently, in the
estimation of models of geography and trade, the gravity model is, once again, holding
center stage (Hummels, 2001, Redding and Venables, 2001, Limao and Venables,
2001). In fact, the gravity model has also become a favored tool to assess ex-post the
trade creating and trade diverting effects associated with preferential trading
arrangements (Frankel, 1997, Soloaga and Winters, 2001).
Along with this renewal in interest, questions have been raised about the proper
formulation of the model (choice of variables) as well as about proper econometric
techniques, especially when the usual cross-country formulation is amended to include a
temporal dimension. Indeed, the discussion about the proper econometric specification
of the gravity model has shown that the conventional cross-section formulation without
the inclusion of country specific effects is misspecified and so introduces a bias in the
assessment of the effects of regional agreements on bilateral trade (e.g., Matyas, 1997
and Soloaga and Winters, 2001). However, it turns out that these specifications, with
3
three specific effects (exporter, importer and time effects) is only a restricted version of
a more general model which allows for country-pair heterogeneity (e.g., Cheng and
Wall, 1999 and Egger and Pfaffermayr, 2000).
In contrast to the traditional cross-section gravity model which includes time invariant
trade impediment measures (e.g. distance, common language dummies, border,
historical and cultural links as in Frankel, 1997), the more general proposed
specification is more adequate since it accounts for any (unobserved) bilateral effect.
Hence, all factors that influence bilateral trade which were partially captured by
regional dummies are now controlled for.
In this paper, I apply this more general specification and show that the predictions of the
effects of regional trade agreements (RTA) in terms of trade creation and trade
diversion are very different according to whether one uses a cross-section or a panel
specification with random bilateral effects (fixed effects eliminating agreements that are
time invariant). In this setting, the potential correlation of some explanatory variables
with the country-pair effects has to be analyzed. I show that the use of the instrumental
method proposed by Hausman and Taylor (1981) is necessary to avoid estimation bias.
Moreover, the selection bias that can appear in an unbalanced sample is tested and
corrected for by the inclusion of a selection rule in the model estimation (as in Guillotin
and Sevestre, 1994).
Section 2 presents the canonical gravity model with the modified cross-section version
used for ex-post evaluations of regional agreements (with the three dummies mentioned
above that have to be included for each RTA according to trade theories). Section 3
4
specifies the alternative panel model with the characteristics proposed above. Finally,
section 4 compares cross-section and panel estimates. To anticipate the main
conclusion, it turns out that the panel estimates yield more convincing estimates which
also suggest that, globally, RTAs generated larger increases in trade among members
than predicted with cross-section estimates. Section 5 concludes.
2. The gravity model as an ex-post method to assess regional agreements
2.1 The standard gravity model
Although several models yield a gravity-type equation, in a framework that emphasizes
aggregate trade, it is convenient to derive the gravity equation from a perfect-
competition H-O type model under the assumption of complete specialization at the
country level, along with product differentiation at the country level. Assume then
maximization of a CES utility function (where σ is the common elasticity of
substitution between any pair of countries’ products, σ>0). As shown in appendix A.1,
this yields the standard “generalized” gravity equation:
(1)
γ
θ=
γ
= σ−
σ−
σ−σ−σ−
σ−
∑∑
σ−σ−
1
hi
hh
i
j
ijijWji
1
hi
hh
i
i
Wjiji
Pp)P(
)p(
eY
YY
Pp)P(
)p(
YYYM
11
∀i,j,h=1..n
where γh is the share of country h in world income, iP is the CES price aggregator in
importer country i and pi is the price in the country of destination i facing consumers.
Assume now that the relationship between the price in the country of origin j, pj, and the
country of destination i, pi is given by :
(2) ijijji epp θ=
5
In (2), eij represents the nominal bilateral exchange rate (defined so that an increase in
its value corresponds to a depreciation of i’s currency with respect to j’s currency) and
θij is a barrier-to-trade function between i and j to be developed below.
To obtain an estimable model from equation (1), three issues need to be considered.
First, distance must be measured correctly. Select units of goods so that each country’s
product price, pj, is normalized to unity (and eij=1). Then, as shown by Deardorff
(1998), iP (given by equation A3 in appendix A.1) becomes a “CES index of country i’s
barriers-to-trade factors” as an importer. Hence, if θij is proxied by distance between i
and j, DISTij, we have to introduce, in addition to the variable of absolute distance
between i and j, a variable of average distance of the importing country i from its main
partners, ( iDIST ) to take account of “the relative distance of i from suppliers” as
suggested by the theoretical models of Anderson (1979) and Deardorff (1998)1.
Omitting this variable would have important consequences, in particular in assessing the
effects of RTAs2.
Second, it is crucial to get the best handle possible on what constitutes the ‘barriers-to-
trade’ function, θij which are usually proxied either by distance, DISTij between trading
partners (and the presence of a common border or language), or sometimes by the
cif/fob price ratio3. Because recent studies have shown that these variables are not the
1 Polak (1996) emphasizes that if one doesn’t use a measure of the average distance between a country and its main
partners as well as absolute distance, one will underestimate trade between faraway countries. 2 If relative distance is not taken into account, the dummy supposed to reflect trade between members of an
agreement will capture the bias. 3 Baier and Bergstrand (2001) use the CIF/FOB ratio to model transport costs, but their study only deals with OECD
countries which have better data. For a discussion about the problems associated with the use of CIF/FOB data see
Hummels (1999) and Limao and Venables (2001).
6
only determinants, we model the barrier-to-trade function, between countries i and j, as
follows4 :
(3) ( ) ( ) ( ) [ ]j6i5ij4321 EELjiijij eININDIST δ+δ+δδδδ=θ
with expected signs on coefficients in parenthesis:
DISTij : distance between the countries i and j (δ1>0);
Lij : takes the value 1 if i and j share a common border, otherwise 0 (δ4<0);
Ei(j) : takes the value 1 if the country i (j) is landlocked; otherwise 0 (δ5>0, δ6>0);
INi (j) : level of infrastructure of the country i (j), computed as an average of the density
of road, railway and the number of telephone lines per capita (δ2<0, δ3<0).
Third, in a sample with countries that have large differences in income per capita, it is
customary to abandon the homothetic utility function and allow Engel effects which
implies including per capita income in the importing country and hence population Ni.
On the supply side, it is reasonable to assume that supply will be driven by factor
endowment differences. Following tradition, we use income per capita as proxy so that
population in the exporting country, Nj is introduced in the model (e.g., Bergstrand,
1989, Frankel, 1997 or Soloaga and Winters, 2001).
Hence, after taking into account the modifications discussed above, the reduced form of
the model is, after substitution of (3) in (1):
(5) lnMij = β0 +β1 lnYi +β2 lnYj +β3 lnNi +β4 lnNj +β5 lnDISTij +β6 ln iDIST +β7 Lij
+β8 Ei +β9 ln INi+β10 Ej +β11 ln INj +ηij
where Yw is absorbed in the constant term, and with expected signs:
β1>0, β2>0, β3<0, β4<0, β5=-σ.δ1<0, β6>0, β7=-σ.δ4>0, β8=-σ.δ5<0, β9=-σ.δ2>0,
4 e.g. Limao and Venables (2001).
7
β10=-σ.δ6<0, β11=-σ.δ3>0, and ηij the error term.
2.2 The gravity model for ex-post assessment of regional trade agreements
First used by Aitken (1973) as an ex-post assessment for the EEC, the gravity model
seems well-defined for this issue for two reasons. First, arguably, the model represents a
relevant counterfactual (or anti-monde) to isolate the effects of an RTA. If the sample of
countries is appropriately selected, the gravity equation then suggests a “normal” level
of bilateral trade for the sample. Then, dummy variables can be used to capture the
“atypical” levels resulting from an RTA.
Second, thanks to the correct introduction of dummy variables in the model, one can
isolate trade creation (TC) and trade diversion (TD) effects of an RTA.
In a Vinerian world following an RTA, TC and TD will be reflected in trade flows as
follows :(i) under pure TC intra-regional trade increases and imports from the ROW
remains unchanged; (ii) under pure TD, the increase in intra-regional trade is entirely
offset by a corresponding decrease of imports from the ROW; (iii) if there is both TC
and TD, intra-regional trade increases more than imports from the ROW decrease.
Because of second-best considerations, identification of TD and TC does not allow
inference about the welfare consequences of an RTA for members. Finally, for non-
members, because under plausible assumption about the anti-monde a necessary
condition for their welfare to increase is that the volume of their imports increases once
the RTA has been established (see Winters, 1997), one should include measure the
change in volume of exports from members to non-members (an increase signifying an
improvement in welfare for non-members).
8
Therefore, the correct ex-post assessment of an RTA on the volume of trade should
include the following dummy variables (associated coefficients in parenthesis)5 :
(i) DI (αI) =1 if both partners belong to the same RTA [zero otherwise] (captures intra-
bloc trade);
(ii) DM (αM) =1 if importing country i belongs to the RTA and exporting country j, to
the ROW [zero otherwise] (captures bloc imports from the ROW);
(iii) DX (αX) =1 if exporting country j belongs to the RTA and importing country i to the
ROW [zero otherwise] (captures bloc exports to the ROW).
Suppose that αI >0 which corresponds to more intra-bloc trade than predicted by the
reference (αI<0 corresponding to an RTA between complementary economies) which
can be in substitution to domestic production or to exports from the ROW. Hence to
conclude on whether this corresponds to TC or TD, one needs to examine the signs of
the coefficients αM and αX. Then, αI>0 along with a lower propensity to import from
the ROW (αM<0) indicates TD, and if the increase in intra-regional trade is entirely
offset by a decrease in regional imports from the ROW, we have pure TD. If intra-
regional trade increases more than imports from the ROW decrease, there is both TC
and TD. And with αI>0 and αM≥0, there is pure TC. Finally, comparing αI and αX can
lead to inferences about welfare for non-members. For example, (αI>0,αX<0) would
indicate a dominant “export diversion” and hence a decrease in welfare for non-
members.
5 This is the specification in Endoh (1999) and Soloaga and Winters (2001). Others assessing the effects
of RTAs (e.g., Bayoumi and Eichengreen, 1997, Frankel, 1997, Krueger, 1999) have not included enough
dummy variables to distinguish between exports and imports, and hence fail to isolate TD and TC effects.
9
To summarize, following an RTA, [αI>0 and αM≥0 (αX≥0)] indicates pure TC in terms
of imports (exports) and [αI>0 and αM<0 (αX<0)], indicates TD in terms of imports
(exports).
3. Data and estimation
The model is estimated with data for 130 countries over the period 1962-96. Trade data
are from UN COMTRADE (bilateral imports in current dollars). The dependent variable
is total bilateral imports and is deflated by a world import price index taken from
International Financial Statistics (IFS). Data sources for the explanatory variables along
with data transformations are presented in appendix A.3. Once the missing values are
taken out6, the sample covers 130 countries (a list of the countries in the sample is
presented in appendix A.4). There are thus 240 691 observations for 14 387 pairs of
countries.
3.1 Panel specification
The usefulness of the gravity model to assess RTAs rests upon a plausible estimation of
the anti-monde. It has been observed repeatedly (see Polak, 1996, Matyas, 1997,
Bayoumi and Eichengreen, 1997) that regional dummy variables in cross-country
estimates capture everything specific to the importing or exporting countries not
captured by the variables included in the equation that influence the level of trade (e.g.
6 The countries which do not declare their imports from a partner or which do not import from this partner are
identified in the same way, with a missing value. Hence, our data are not censored at zero. The actual number of
observations (240 691) represent around 50% of potential number. The selection bias which can exist is tested and
corrected by the inclusion of the selection rule in the model estimation in section 4.
10
historical, cultural, ethnic, political or geographical factors)7 which is troublesome since
the dummy variables should really isolate TD and TC effects. Not taking into account of
countries’ heterogeneity or of the pair of countries in bilateral trade relations introduces
a bias. By contrast, a panel data method enables one to identify the specific effects of
the pair of countries and to isolate them. The inclusion of this bilateral term, αij, specific
to each pair of countries and common to each year (and different according to the
direction of trade: αij ≠ αji), is more general than dummies capturing specific elements of
trade such as common language or cultural similarity (see Cheng and Wall, 1999, Egger
and Pfaffermayr, 2000).
So the previous model is specified in panel as:
(6) lnMijt = α0 + αt + αij +β1 lnYit +β2 lnYjt +β3 lnNit +β4 lnNjt +β5 lnDISTij
+β6 ln iDIST +β7 Lij +β8 Ei +β9 ln INit+β10 Ej +β11 ln INjt +β12 lnRERijt +η’ijt
α0 : effect common to all years and pairs of countries (constant);
αt : effect specific to year t but common to all the pairs of countries8;
αij : effect specific to each pair of countries and common to all the years.
Note the introduction of the bilateral real exchange rate (RERijt) in (6). In a model with
panel data that spans a long time period (here 35 years), it is essential to capture the
evolution of competitiveness. Given our definition of eij in equation (1), an increase of
the RER reflects a depreciation of the importing country’s currency against that of the
exporting country which should reduce imports (hence one would expect β12<0).
7 If these factors are also correlated with gravity variables (GDP, populations, distance), estimations which do not
include them will have an endogeneity bias, because the omitted variables are correlated with the level of bilateral
trade and with the explanatory variables (see below). 8 These time dummies capture common shocks such as the evolution of oil prices over the period or Yw in (1).
11
3.2 Econometric method
Since a fixed effects model is inadequate (the within transformation eliminates time-
invariant variables), bilateral effects are modeled as random variables. Time effects are
captured by yearly dummies that capture common shocks (e.g. oil price changes). In the
absence of correlation between the explanatory variables and the specific bilateral
effects, the GLS estimation provides consistent estimates of the coefficients. However,
variables like GDP or infrastructure may be correlated with bilateral specific effects9..
The usual way to deal with this issue is to consider an instrumental variables estimation
such as that proposed by Hausman and Taylor (1981) (see appendix A.5 for the
implementation of this method), though here it is adapted to the case of an unbalanced
sample according to the method proposed by Guillotin and Sevestre (1994).
A Hausman-Taylor test of over-identification, based on the comparison of the
Hausman-Taylor estimator and the Within one, must be carried out (see appendix A.5).
If the null hypothesis cannot be rejected, the instruments are legitimate (in the sense of
no bias due to a correlation between specific bilateral effects and the explanatory
variables), and the Hausman-Taylor estimator (HT) is the most efficient estimator.
9 The Hausman test (1978) allows us to control for the presence of correlation between explanatory variables and specific bilateral effects.
12
3.3 Endogenity of explanatory variables and sample selection bias
Before proceeding to the evaluation of the effects of RTA (which are, in this section,
captured by the bilateral specific effect), I check first for endogeneity of explanatory
variables.10 Results are reported in table 1.
Column 1 in table 1 reports estimates from the Within equation which treats the
bilateral specific effects as fixed, thereby giving unbiased parameter estimates for time-
varying variables. All these coefficients are significant at a 99% level and have the
expected sign. The fit is good (R2=0.87) and the specific bilateral and time effects
introduced in the model are strongly significant (as showed by the Fisher tests).
Next come the results from estimating the error component model (GLS) which differ
markedly from the Within estimation. The Hausman test, based on differences between
Within and GLS estimators, reveals a χ²7 = 462.07, which is significant at 99%. Hence,
this test rejects the null hypothesis according to which there would be no correlation
between the bilateral specific effects and the explanatory variables. The GLS estimator
is thus biased, and the use of the Hausman-Taylor method justified.
For sensitivity analysis, four regressions are estimated with the Hausman-Taylor
method. The over-identification test indicates for each regression if the instruments are
legitimate or if an additional source of correlation between specific effects and
explanatory variables exists (in the case of a significant test statistic).
10 Because the dataset covers a long time span, some series may contain a unit root and thus the estimates in the table 1 may be spurious. So, a Levine and Lin (1993) unit root test has been applied to the series for GDP, population and bilateral import. This test rejects, for all series, the null of a unit root.
13
Table 1 : Results of the estimates of the gravity equation on panel data.
Mijt Variables Within GLS HT I a) HT II b) HT III c) HT IV d)
ln Yit 1.03** 0.79** 1.00** 1.10** 1.19** 1.03** (49.8) (78.4) (60.4) (74.5) (74.4) (68.4)
ln Yjt 1.11** 1.12** 1.18** 1.17** 1.13** 1.11** (59.8) (129.8) (109.4) (107.0) (98.4) (80.2)
ln Nit 0.19** 0.017 -0.088** 0.20** 0.11** 0.20** (5.15) (1.5) (-6.6) (8.9) (4.7) (7.7)
ln Njt -0.65** -0.19** -0.25** -0.63** -0.65** -0.65** (-24.8) (-17.4) (-19.5) (-24.1) (-24.6) (-24.6)
ln DISTij - -1.14** -1.17** -1.19** -2.09** -1.19** (-59.0) (-51.1) (-48.4) (-14.3) (-45.3)
ln iDIST - -0.66** 0.26** 1.13** 2.02** 1.10** (-26.4) (5.1) (17.1) (22.2) (15.8)
Lij - 0.68** 1.14** 1.01** 0.77* 1.04** (9.6) (10.4) (8.8) (2.5) (8.5)
Ei - -0.27** -0.18** -0.02 -0.10 -0.02 (-5.4) (-3.1) (-0.4) (-1.7) (-0.3)
Ei - -0.47** -0.41** -0.58** -0.59** -0.56** (-10.4) (-8.1) (-11.2) (-10.6) (-9.6)
ln INit 0.04** 0.04** 0.04** 0.05** 0.07** 0.04** (5.3) (6.3) (6.8) (6.3) (9.9) (6.0)
ln INjt 0.03** 0.02** 0.01** 0.03** 0.03** 0.03** (5.9) (5.6) (3.1) (6.0) (5.9) (6.5)
ln RERijt -0.006** -0.005** -0.004** -0.006* -0.006** -0.006** (-5.7) (-3.6) (-2.7) (-4.1) (-4.4) (-4.2)
Number of obs (NT) 240 691 240 691 240 691 240 691 240 691 240 691 Number of bilateral (N) 14 387 14 387 14 387 14 387 14 387 14 387
R² e) 0.87 0.63 0.62 0.61 0.61 0.63 Theta (mean) - 0.82 0.83 0.83 0.83 0.84
Bilateral fixed effect 37.09** - - - - F(14386,226263)
Time fixed effect 59.97** 161.31** 98.61** 131.58** 144.44** 156.31** F(34,226263) F(34,240644) F(34,240644) F(34,240644) F(34,240644) F(34,240644)
Hausman test W vs. GLS f) - 469.07** - - - chi-2(Kw) chi-2 (7)
Hausman test HT vs. GLS g) - - 961.27** 1184.41** 1480.83** 973.53**chi-2(K) chi-2(12) chi-2(12) chi-2(12) chi-2(12)
Test of over-identification h) - - 243.91** 13.21** 110.33** 0.39 chi-2 (k1-g2) chi-2(5) chi-2(3) chi-2(1) chi-2(1)
** and * significant at 99% and 95% respectively (t-student is presented under the correspondent coefficient). The time dummy variables and the constant are not reported in order to save space. a) HT I : endogenous variables = lnYit et lnYjt, k1-g2=5. b) HT II : endogenous variables = lnYit, lnYjt, lnNit et lnNjt, k1-g2=3. c) HT III : endogenous variables =lnYit, lnYjt, lnNit, lnNjt, lnDISTij et ln iDIST , k1-g2=1. d) HT IV : endogenous variables = lnYit, lnYjt, lnNit, lnNjt, lnINit et lnINjt, k1-g2=1. e) Calculated, for GLS and HT, from [1-Sum of Square Residuals] / [Total Sum of Squares] on the transformed
model. Note that the impact of random specific effects are not in the R2 but are part of residuals. f) This test is applied to the differences between the Within and GLS estimators, without taking into account the
coefficients of time effects. If we take them into account, the result is: chi-2(41)= 853.61** g) Hausman test applied to the differences between GLS and HT estimators, without time effects. h) Hausman test applied to the differences between Within and HT estimators, without time effects.
Cf. a), b) c) and d) for information on k1-g2.
14
The first estimation, labeled HT I in table 1, considers only the GDP variables (Yit and
Yjt) as endogenous. The results point out that these variables are actually correlated with
the specific effects: the Hausman test, which compares HT I to GLS, confirms that the
instrumentation has improved the model11 (the hypothesis of exogeneity of GDP
variables is rejected). However, the over-identification test rejects the hypothesis
according to which there would be no more correlation between explanatory variables
and bilateral effects (χ²5 = 243.91). Hence, only a part of the initial bias has been
corrected.
A second source of correlation can come from the population variables. Equation HT II
takes these two variables (and the GDP variables) as endogenous. The corresponding
tests for this equation lead us to conclude that once again, the model has been improved
but the difference with the Within estimation is still significant.
A third source of endogeneity can be due to the variables of infrastructure12. Their
instrumentation, in addition to those of income and population, improves the model and
the over-identification test indicates that the hypothesis of legitimacy of the instruments
used cannot be rejected. As the identification condition is verified (see appendix A.5),
the Hausman-Taylor estimator is convergent and more efficient than the Within
estimator13.
11 Guillotin and Sevestre (1994) recommend comparing the HT estimator, denoted βHT, to the GLS estimator, denoted
βMCQG. It is exactly the same principle as for earlier tests presented in appendix A.5. We compute the Hausman
statistic test and the number obtained is compared to the critical value of χK2 (K is the dimension of the coefficients’
vector βMCQG). If the null H0 is rejected, we can conclude that the instrumented model gives better estimations then
the GLS model (without any instrumentation). Thus, the instrumented variables are actually endogenous. 12 I also test for the correlation of distance variables with bilateral effects in the equation HT III. However this
equation does not improve the model HT II. 13 According to the Barghava and al. Durbin Watson test (1982), modified to the unbalanced panel, the HT IV
residuals are no autocorrelated AR(1): there is no systematic difference between observed and predicted trade flows.
Hence, the HT IV estimator is efficient and the over-identification test is appropriate (e.g. Egger 2002).
15
All these coefficients are significant at a 99% level (except for Ei) and have the
expected sign. Import volume of i from j increases with GDP and coefficients are close
to unity as suggested by the theory and as reported by, for instance, Aitken (1973),
Soloaga and Winters (2001) and Egger and Pfaffermayr (2000). The population variable
has the expected negative sign for the exporting country (capturing the fact that larger
countries trade less) but has a positive sign for the importing country (as in e.g., Soloaga
and Winters, 2001, and Egger and Pfaffermayr, 2000). The elasticity of bilateral trade to
distance is superior to unity (-1.19) and the volume of trade increases with the level of
infrastructure of each country, as in Limao and Venables (2001). Sharing a land border
allows countries to trade 2.8 times more than expected from the gravity equation
(=exp(1.04)). Likewise, imports from a country without direct access to the sea are
43% lower. Finally, a real depreciation of i with respect to j lowers i’s imports from j.
A last potential estimation bias must be considered: the unbalanced sample can be
subject to a non-ignorable selection rule, i.e. that the probability of a pair of countries
being included in the sample is not independent of model error, and in particular to the
unobserved bilateral effects. In this case, the selection bias can be tested and corrected
by the inclusion of the selection rule in the model estimation. I use a method proposed
by Nijman and Verbeek (1992): which approximates the Heckman correction term14, by
adding variables which reflect the individual’s patterns in terms of presence in the
sample to the model. So HT IV is estimated again including the following additional
variables: (i) PRES: number of years of presence of the couple ij’s in the sample;(ii)
14 I use an alternative method because the generalization of the Heckman two-steps method (1979) to panel data and
random effects model is too difficult (Guillotin and Sevestre 1994, p.127).
16
DD: dummy that takes the value 1 if ij is observed during the entire period, 0
otherwise;(iii) PAt: dummy that takes the value 1 if ij was present in t-1 (PA0=0).
Results from this estimation are reported in appendix A.6 and are compared to equation
HT IV (table 1). In the first column, with only the variable PRES considered, the
conclusions of the previous estimations are not modified, even if the coefficient of
PRES is statistically different from zero. The following regressions (columns 2 and 3)
show that the variables DD and PAt have a positive and significant coefficient: all other
things equal, pairs of countries which have at least two years of consecutive available
data (and a fortiori if they are present over the entire period) have more bilateral trade
than pairs of countries with interruption in their data. These three variables will be
systematically introduced in future regressions, in order to avoid the selection bias in
the coefficients of regional dummies.
4. Application to the assessment of the effects of regional trade agreements
Following the specification check, the three dummy variables discussed above were
introduced in the model to detect TD and TC for a selection of RTAs (EU, ANDEAN,
NAFTA, CACM, MERCOSUR, ASEAN, EFTA, LAIA). To save space, detailed
comments are only reported for three well-known RTAs: EU, NAFTA, and
MERCOSUR, the EU being included in spite of lack of data for years prior to the
agreement because it is the best-known and most studied RTA. Average effects over the
sample period are reported first, then effects over time to look for break points around
the important dates of the agreements.
17
4.1 Average effects over the period 1962-1996
Table 2 reports the coefficients for dummy variables for two sets of regressions, one in
cross-section (corresponding to most uses of the gravity model for ex-post assessments
of RTAs), yielding 35 separate regressions (one for each year), the other with the panel
specification of section 3. All results are presented in appendix A.7.
Table 2 : Results for regional dummies over 1962-96.
Mijt Variables Panel
(HT IV) Cross-section
(average coefficients) a)
EUintra 0.291* -0.215 EUimports 0.225** 0.797 EUexports 0.375** 0.746 MERCOSURintra -0.275 -0.432 MERCOSURimports -1.041** 0.017 MERCOSURexports -0.130** 0.088 NAFTAintra -0.063 0.754 NAFTAimports -0.478** 0.253 NAFTAexports 0.009 0.011 ** and * significant at 99% and 95% respectively for Panel estimation. a) For each variable, this is the average of the 35 coefficients estimated per year from 1962 to 1996.
Generally, the significance of the coefficients is greater for the panel specification, with
coefficients of the same sign when they are significant in both specifications.
These result give the average impact of each RTA over 1962-96. However, relevant
inferences about TD and TC require inspection of the evolution of these coefficients
over time and around the period when RTAs go into effect which can be done by
breaking down estimation into subperiods. To this effect, I break down regional dummy
variables into two-year periods with these variables introduced in the estimating
equation instead of the global regional dummies.15
15 Coefficient estimates for explanatory variables are identical under this procedure because for each
agreement, the addition of the new variables introduced is equal to the former aggregate dummy variable.
18
4.2 Evolution of the effects during the RA’s existence
Because the results are self-explanatory from inspection of figure 1 to 3, I comment the
EU results and give only an overall interpretation for NAFTA and MERCOSUR along
with a summary for other RTAs16. All the evolutions commented in this section are
significant one17.
Start with the cross-section analysis (figure 1a) for the EU which displays a negative
trend in intra-EU trade until 1980 before turning positive with the propensity to export
to the ROW declining over the period suggesting exports TD but no evidence of import-
TD, a result similar to Soloaga and Winters (2001) obtained using the same estimation
method.
16 The evolution of the estimated coefficients are represented in appendix A.8 for cross-section results and
for panel ones for the other RTAs. 17 The consecutive coefficients are tested significantly different.
19
Figure 1: evolution of EU dummies over 1962-1996 (αI, αM and αX)
Figure 1a: evolution of the EU dummies estimated in cross-section
-3
-2
-1
0
1
2
3
1962 1966 1970 1974 1978 1982 1986 1990 1994
EU-intra EU-imports EU-exports
Figure 1b: evolution of the EU dummies estimated in panel
-3
-2
-1
0
1
2
3
1962 1966 1970 1974 1978 1982 1986 1990 1994
EU-intra EU-imports EU-exports
First enlargment Second enlargment First enlargment Second enlargment
By contrast, panel estimates (figure 1b) suggest three rather distinct periods in terms of
TC and TD. From 1967 to 1973, intra-trade decreases somewhat surprisingly without
clear tendencies for trade with ROW. However, following the first (and second)
enlargements, the models predicts a significant positive trend in intra-trade (αI increases
and turns positive in 1984, the pattern continuing with the deep integration following
the EC-92 programme). In parallel, there is first a stagnation of imports of members
from the ROW until 1985 and then a negative trend (αM became negative in 1990).
Hence, the model suggests that, if the first enlargement of the EU (from six to nine
members in 1974) resulted in a pure TC, the second enlargement (with Spain and
Portugal in 1986 and subsequent deep integration) presents sign of significant TD, in
terms of imports and exports. Note however that deep integration in the form of reduced
technical barriers to trade, even if discriminatory, cannot give rise to welfare reduction
for RTA members. These results are quite different from Bayoumi and Eichengreen
(1997) who found a TD after the first enlargement and TC after the second, but are
closer to those of Frankel (1997), Krueger (1999) or Soloaga and Winters (2001).
20
Figure 2: evolution of MERCOSUR dummies over 1962-1996 (αI, αM and αX)
Figure 2a: evolution of the MERCO SUR dummies estimated in cross-section
-3
-2
-1
0
1
2
3
1962 1966 1970 1974 1978 1982 1986 1990 1994
MERCOSUR-intra MERCOSUR-importsMERCOSUR-exports
Figure 2b: evolution of MERCO SUR dummies estimated in panel
-3
-2
-1
0
1
2
3
1962 1966 1970 1974 1978 1982 1986 1990 1994
MERCOSUR-intra MERCOSUR-importsMERCOSUR-exports
Asuncion Treaty
Protocols Arg-Brazil
Asuncion TreatyProtocols Arg-Brazil
Figure 3: evolution of NAFTA dummies over 1962-1996 (αI, αM and αX)
Figure 3a: evolution of NAFTA dummies estimated in cross-section
-3
-2
-1
0
1
2
3
1962 1966 1970 1974 1978 1982 1986 1990 1994
NAFTA-intra NAFTA-importsNAFTA-exports
Figure 3b: evolution of NAFTA dummuies estimated in panel
-3
-2
-1
0
1
2
3
1962 1966 1970 1974 1978 1982 1986 1990 1994
NAFTA-intra NAFTA-importsNAFTA-exports
NAFTA signed NAFTA signed
Mexico’s unilateral trade liberalisation
Mexico’s unilateral trade liberalisation
Comparing the results from both estimation methods is even more striking in the cases
of MERCOSUR and NAFTA. Here, the cross-section estimates show largely
unexplainable volatility throughout the time-period whereas the panel estimates capture
much more clearly the expected effects of an RTA around the time of announcement or
implementation: an increase in intra-trade and a decrease in imports from the ROW. The
difference in patterns is particularly striking for NAFTA which reveals largely
insignificant dummies until the first trade policy reforms in Mexico, and the
announcement of NAFTA negotiations. As to MERCOSUR, panel estimates capture
both the increase in intra-trade and the diversion of import from the ROW captured in
the more disaggregated analysis in Yeats (1998). At the same time, there is some
32
evidence of an increase of the exports for NAFTA and MERCOSUR to the ROW
(which probably reflects the opening up of the countries to the world as the same time
as they were forming the RTA). Clearly, the panel estimates reveal a more plausible
pattern than the cross-section estimates.
This pattern of import (and sometimes export) TD was also found for other RTAs
reported in appendix A.8. For example, in the case of the ANDEAN accord, the model
finds import-TD over the period 1969-79, over the period 1962-77 for the CACM, and
over the period 1968-1980 for the LAIA. Concurrently, over the same period, an export-
TD is observed for the ANDEAN, whereas there is some evidence of an increase of the
propensity to export towards ROW for CACM. No clear patterns emerged for EFTA,
while ASEAN and LAIA are the only examples of pure TC over the period.
5. Conclusions
This paper has paid particular attention to the specification and the estimation of the
gravity model to correct for biases present in previous studies. The panel estimation
with bilateral specific random effects was revealed to be statistically justified after
correction for endogeneity of the income, size and infrastructure variables. Moreover,
dummies were introduced to take into account the selection rule of the sample.
Arguably, these modifications lead to a better formulation of the anti-monde against
which one assesses the trade performance of RTAs.
Comparison of panel estimates with the more usual cross-section estimates revealed a
far more plausible pattern of trade effects associated with RTAs as evidenced by
32
examination of three well-studied RTAs: EU, MERCOSUR and NAFTA. In general,
the results in this study, covering eight RTAs, show that most of them resulted in an
increase in intra-regional trade beyond levels predicted by the anti-monde reference,
often coupled with a reduction in imports from the ROW, and at times coupled with a
reduction in exports to the ROW, suggesting evidence of trade diversion.
Appendices
A.1 : Derivation of the gravity model
As in Deardorff (1998), assume each country i is specialized in a single commodity, with a
representative consumer maximizing a homothetic utility function:
( ) 11
jijji CbU
−σσ
σ−σ
= ∑ (A1)
where σ is the common elasticity of substitution between any pair of countries’ products subject
(σ>0), and bj=bi, ∀i,j guarantees symmetry and a single price for each product variety. Product
differentiation is at the national level (rather than at the firm level as in the monopolistic
competition version), and CES preferences (rather than Cobb-Douglas) implies that bilateral
trade decreases with distance. Each consumer Maximization of (A1) subject to the budget
constraint Yi=pixi (with xi the production of country i) gives:
i1
ii
ji
ji YPpbp
1Cσ−
= (A2)
where ( )σ−
σ−
= ∑1/1
j1iji pbP (A3)
is the CES price aggregator in country i associated with the minimization of expenditures in the
utility maximization problem and pi is the price in the country of destination i facing consumers.
Assume that the relationship between the price in the country of origin j, pj, and the country of
destination i, pi is given by :
ijijji epp θ= (A4)
In (A4), eij represents the nominal bilateral exchange rate and θij the barrier-to-trade function
between i and j. This term is usually proxy by the distance between the two countries.
32
To get the standard gravity-based model, assume balanced trade and let γj=Yj/YW be the share of
country j in world income, YW. Expenditures of all countries i on the good produced in j are
. Then, Yjii
iCp∑ j= jii
iCp∑ and substituting the value of Cji from (2) into this expression gives:
11
ii
iijj
Ppb
−σ−
γγ= ∑ (A5)
Substituting (A5) into (A2), the volume of imports of country i from j is given by:
γ
θ=
γ
= σ−
σ−
σ−σ−σ−
σ−
∑∑
σ−σ−
1
h
i
hh
i
j
ijijWji
1
h
i
hh
i
i
Wji
ji
Pp)P(
)p(
eY
YY
Pp)P(
)p(
YYYM
11
∀i,j,h=1..n (A6)
The intensity of trade between two countries is a function of their respective size and that it is a
decreasing function of the extent of barriers to trade θij.
To simplify this, first select units of goods so that each country’s product price, pj, is normalized
to unity (and eij=1). Then, as shown by Deardorff, iP (given by A3) becomes a CES index of
country i’s barriers-to-trade factors as an importer. Using Deardorff’s notation, the average
barrier-to-trade from suppliers, δ , is given by: Si
( )( )σ−
σ−
θ=δ ∑
1/1
j
1ijjS
i b (A7)
Substituting (A7) into (A6) gives expression:
δθγ
δ
θ=
∑σ−
σ−
σ−
h
1
Sh
hjh
1
Si
ijW
jiji
1
YYYM (A8)
32
A.2 : Definition of the regional agreements studied UE EFTA NAFTA LAIA CACM ANDEAN MERCO
SUR ASEAN
1962 1957(EEC) 1960 1960 (LAFTA) 1960 France Austria Argentina Costa Rica Germany Denmark Bolivia El Salvador Belgium Norway Brazil Guatemala Italy Portugal Chile Honduras Luxembourg Sweden Colombia Nicaragua 1964 Netherlands Switzerland Ecuador UK Mexico Finland Paraguay 1966 Peru Uruguay 1967 Venezuela Indonesia1969 1969 Singapore Bolivia Philippines
Chile Malaysia
1970 Colombia Thailand1973 1973(EEC) Austria Ecuador France Iceland(70) Peru 1975 Germany Norway Venezuela(73) Belgium Portugal 1976 Italy Sweden Bolivia Luxembourg Switzerland Colombia 1980 Netherlands Finland 1980 (LAIA) Ecuador UK Argentina Peru Denmark Bolivia Venezuela 1984 Ireland Brazil Greece (81) 1985 Chile Spain (86) Austria Colombia Portugal (86) Iceland(70) Ecuador Austria (95) Norway Mexico Finland (95) Sweden Paraguay Sweden (95) Switzerland Peru Finland Uruguay 1991 Venezuela 1991 Argentina 1992 1992 Brazil 1992 Canada 1992 Uruguay Indonesia Mexico Bolivia Paraguay Singapore1994
USA Colombia Philippines
Ecuador Malaysia Venezuela Thailand 1995 Norway Switzerland 1996
Liechtenstein (91)
Bilateral trade of Liechtenstein and Switzerland is not desegregated in this data set (as for Belgium and Luxembourg).
32
A.3 : Sources and data definition
Mijt : COMTRADE, total bilateral imports of country i from country j at time t. This variable is
in current dollar so it has been divided by an index of the unit value of imports, which is
taken from IMF, to obtain a real flow of trade.
Yi(j)t : CD-ROM WDI, World Bank 1999, GDP of country i at time t in constant dollar 1995.
Ni(j)t : CD-ROM WDI, World Bank 1999, total population of country i at time t.
DISTij : Data for distance are extracted from the software developed by the company CVN. The
distance is measured in kilometers between the main city of the country i and that of
country j. Most of the time, the main city is the capital city, but for some countries the main
economic city is considered. The distance calculated by this software is orthodromic, that
is, it takes into account the sphericity of Earth. More precisely, ‘the distance between two
points A and B is measured by the arc of the circle subtended by the chord [AB]’ (see
HAINRY, «Jeux Mathématiques et Logiques – Orthodromie et Loxodromie »).
Lij : Dummy equal to one if the countries i and j share a common land border, 0 otherwise.
Ei(j) : Dummy equal to one if the country i is landlocked (i.e. do not have a direct access to the
sea), 0 otherwise.
INi(j)t : This index is built using 4 variables from the database constructed by Canning (1996):
the number of kilometer of roads, of paved roads, of railways, and the number of telephone
sets/lines per capita of country i (j) at time t. The first three variables are divided by the land
area (WB, 1999) to obtain a density. Thus, each variable obtained is normalized to have a
same mean equal to one. An arithmetic average is then calculated over the four variables,
for each country and each year, without taking into account the missing values (a similar
computation is presented by Limao and Venables 2001). As the final year of the data set is
1995, an extrapolation had to be made to cover the year 1996.
iDIST : average distance of country i to exporter partners, weighted by exporters’ GDP share in
world GDP (“remoteness” of country i). The ten main trade partners are identified for each
country according to bilateral flows averaged over 1980-96 (in COMTRADE). For the
weights, we used 1990’s GDP (WB, 1999). Hence, This variable is specific to each country
and is not time variant.
RERijt : We extract from the IFS data set the nominal exchange rate for each country against
US dollar (NERi/$, country i’s currency value of 1 US$), and the consumption price index for
country i (CPIi), for each year from 1962 to 1996. If the CPI is not available for a country,
we consider the GDP deflator of the country. The bilateral real exchange rate (RER) is
computed as following: RER i/j = (CPIj) / (CPIi) . (NERi/$ / NERj/$ ), where i is the importing
country and j the exporting one. For each pair of countries, we specify the RER such as its
mean over the period is zero.
32
A.4 : Countries in the sample.
OECD Sub-Saharan Africa
Latin America and the Caribbean
Asia and the Pacific
Others
Australia Angola Argentina Bangladesh AlbaniaAustria South Africa* Bahamas Brunei Armenia
Burundi Barbados Bhutan Azerbaijan Belgium + Luxembourg Benin Belize China Bulgaria
Canada Burkina Faso Bolivia Fiji Belarus Germany Central African Rep. Brazil Hong Kong Czech Rep. Denmark Ivory Coast Chile Indonesia Algeria
Spain Cameroon Colombia India Saudi Arabia Finland Congo Costa Rica Cambodia Egypt France Comoros Dominican Rep. Lao PDR Estonia
United Kingdom Cape Verde Dominica Macao Georgia Ireland Djibouti Ecuador Mongolia Greece Iceland Ethiopia + Eritrea Grenada Malaysia
Italy Gabon Guatemala Nepal Bosnia and
Herzegovina Japan Ghana Guyana Pakistan Hungary
Korea, Rep. Guinea Honduras Philippines Iran United States Guinea-Bissau Haiti Papua New Guinea Israel Netherlands Gambia Jamaica Singapore Jordan
Norway Equatorial Guinea Mexico Salomon Islands Kazakstan New Zealand Kenya Nicaragua Thailand Kyrgyz Rep.
Portugal Madagascar Panama Vietnam Kuwait Sweden Mali Peru Western Samoa Lithuania
Mozambique Paraguay Sri Lanka Latvia Switzerland + Liechtenstein Mauritania El Salvador Tonga Macedonia
Mauritius Suriname Kiribati Morocco Malawi Trinidad and Tobago Vanuatu Malta Niger Uruguay Oman Nigeria St. Vincent and Poland Rwanda The Grenadines Romania Sudan Venezuela Russian Federation Senegal St. Lucia Slovenia Sierra Leone Antigua and Slovak Rep. Sao Tomé and Principe Barbuda Syrian Rep. Seychelles St. Kitts and Tajikistan Somalia Nevis Turkmenistan Chad Tunisia Togo Turkey Tanzania Ukraine Uganda Uzbekistan Zaire Zambia Zimbabwe
Countries written in italic are not available as reporter countries in COMTRADE (only as partners). * South Africa includes bilateral trade of the group of countries: South Africa + Lesotho + Botswana + Namibia + Swaziland.
32
A.5: the Hausman and Taylor (1981) method.
Let us consider:
Mijt= Xijtβ + Zijδ+ uijt with uijt = αij + νijt (A.9)
With Xijt = [lnYit lnYjt lnNit lnNjt lnINit lnINjt lnRERijt]
and Zij= [lnDISTij ln iDIST Lij Ei Ej]
where some explanatory variables of X (variables variant over time) and of Z (time-invariant
variables) are correlated with the specific effects. We suppose that among the variables X and Z,
there exist:
(i) Xijt: k1 (k2) exogenous (endogenous) variables, denoted X1 (X2);
(ii) Zij: g1 (g2) exogenous (endogenous) variables, denoted Z1 (Z2);
If the condition k1 ≥ g2 is satisfied, then the equation is identified18 and (A.9) can be estimated
using [QX1, QX2, PX1, Z1]19 as instruments (see Breusch, Mizon and Schmidt, 1989). The
instruments are then taken within the model. The resulting estimator is consistent but not
efficient, as it does not correct for heteroskedasticity and serial correlation due to the presence
of random bilateral specific effects. Hence, Hausman and Taylor (1981) suggest using this first
round of estimates to compute the variance of the specific effect (σµ²) and the variance of the
error term (σν²). The instrumental variable estimator is then applied to the following transformed
equation:
Yijt – (1-θ) Yij. = [ Xijt – (1-θ)Xij. ] β + θZijδ + θµij + [νijt –(1-θ) νij.]
With20 θ= ( σν2 / Tσµ
2 + σν
2)1/2 (A.10)
A test of over-identification must be carried out. It is based on the comparison of the Hausman-
Taylor estimator, denoted βHT, and the Within estimator (fixed effects model), denoted βw. The
Hausman test statistic is:
[βHT - βW] . [var(βHT) - var(βW)]-1 . [βHT - βW]’ (A.11)
Under the null hypothesis, this test statistic is distributed as a Chi-square (χ2) with k1-g2 degrees
of freedom. If the statistic is inferior to the critical value, then the null can’t be rejected: the
instruments are legitimate21. If k1 > g2, we can also conclude that the Hausman-Taylor estimator
(HT) is the most efficient estimator.
18 If k1 > g2 then the equation is over-identified. 19 Q is the matrix that computes the deviations from individual means. P is the matrix that computes the observation
across time for each individual (pair of countries). 20 Owing to the fact that our sample is unbalanced, we have in fact θij= ( σν
2 / Tijσµ2 + σν
2)1/2 with σµ2 et σν
2 which
are corrected for the bias of heteroskedasticity, specific to the unbalanced sample, according to the method proposed
by Guillotin and Sevestre (1994). The mean value of θij will be systematically presented in the tables of results. 21 Actually, the null hypothesis H0 is that there are no significant difference between the Within estimator and the HT
one. So, under H0, there is no longer bias due to a correlation between specific bilateral effects and explanatory
variables.
32
A.6 : Test and correction of selection bias in equation HT IV
Variables
Mijt
1 2 3ln Yit 1.00** 1.06** 0.98**
(61.1) (65.0) (60.8)
ln Yjt 1.13** 1.15** 1.14** (77.1) (76.9) (89.5)
ln Nit 0.16** 0.19** 0.16** (6.4) (7.3) (6.2)
ln Njt -0.64** -0.69** -0.62** (-21.4) (-26.1) (-22.8)
ln DISTij -1.09** -1.14** -1.17** (-44.8) (-43.8) (-43.9)
ln iDIST 0.96** 1.17** 0.60** (14.3) (17.0) (8.7)
Lij 0.97** 1.04** 0.90** (8.8) (8.7) (7.2)
Ei -0.17 -0.05 -0.14* (-5.2) (-0.8) (-2.3)
Ej -0.54** -0.52** -0.51** (-6.4) (-9.1) (-4.9)
ln INit 0.04** 0.04** 0.04** (5.1) (5.6) (4.6)
ln INjt 0.03** 0.03** 0.03** (7.3) (6.6) (7.1)
ln RERijt -0.006** -0.006** -0.005** (-4.1) (-4.3) (-3.1)
PRES 0.05** - 0.039** (29.2) (18.5)
DD - 0.84** 0.10 (13.5) (1.5)
PAt - 0.49** (49.4)
Number of obs (NT) 240 691 240 691 240 691 Number of bilateral (N) 14 387 14 387 14 387
R² 0.63 0.64 0.65 Theta (mean) 0.83 0.84 0.85
** and * significant at 99% and 95% respectively ( t-student is presented under the correspondent coefficient). The time dummy variables and the constant are not reported in order to save space. The estimation method is one of Hausman-Taylor, with variables Yit, Yjt, Nit, Njt, INit and INjt as endogenous (HT IV).
32
A.7 : Results of the estimation with regional dummies (1962-1996). Mijt
Panel Cross-section Variables Coeff. t Average Coeff. max min
ln Yit 1 033** 62.66 0 769 0.97 0.63 ln Yjt 1.139** 85.15 0.935 1.21 0.69 ln Nit 0.131** 11.38 -0.045 -0.01 -0.12
ln Njt -0.650** -22.69 -0.082 -0.01 -0.20 ln DISTij -1.168** -41.93 -0.971 -0.46 -1.25
ln iDIST 0.751** 16.11 0.136 0.65 -0.71 Lij 0.921** 7.68 1.018 1.58 0.54
Ei -0.161** -3.66 -0.178 -0.01 -0.63 Ej -0.510* -2.09 -0.389 -0.05 -1.07
ln INit 0.037** 4.68 0.157 0.26 0.07 ln INjt 0.031** 7.19 0.067 0.14 0.01
ln RERijt -0.005** -4.12 - - -
PRES 0.039** 19.21 - - - DD 0.099 0.31 - - - PAt 0.494** 49.4 - - -
EU intra 0.291* 1.98 -0.215 0.58 -0.88 EU imports 0.225** 3.16 0.797 1.04 0.04
EU exports 0.375** 5.11 0.746 1.53 0.02
EFTA intra -0.287 -1.56 0.319 0.77 -0.10 EFTA imports -0.075 -0.89 -0.098 0.12 -0.55 EFTA exports -0.932** -11.93 -0.007 0.37 -0.21
ASEAN intra 0.680** 4.18 1.757 2.78 1.22 ASEAN imports -0.513** -5.77 0.458 0.96 0.01
ASEAN exports 0.757** 8.86 0.421 1.15 -0.25
ANDEAN intra 0.772** 4.78 1.049 2.65 -0.25 ANDEAN imports -0.940** -5.07 0.285 1.2 -0.64 ANDEAN exports -0.959** -6.08 -0.022 1.79 -1.37
MERCOSUR intra -0.275 1.54 -0.432 1.01 -1.68 MERCOSUR imports -1.041** -6.76 0.017 0.57 -0.63
MERCOSUR exports -0.130 0.93 0.088 1.06 -1.02
LAIA intra a) 0.360** 4.62 0.327 1.2 -0.77 LAIA imports -1.492** -12.23 -1.073 -0.48 -1.92 LAIA exports -0.357 1.41 -0.359 0.88 -1.23
CACM intra 1.087** 3.91 2.305 3.44 1.19 CACM imports -0.776** -8.30 -0.498 0.06 -0.88
CACM exports -0.127 -1.34 -0.097 0.28 -0.79
NAFTA intra -0.063 -0.48 0.754 2.18 -0.30 NAFTA imports -0.478** -5.96 0.253 0.68 -0.21 NAFTA exports 0.009 0.07 0.011 0.76 -0.65
Number of obs (NT) 240 691 7 265 9 362 5 819 Number of bilateral (N) 14 387 - - - R² 0.66 0.64 0.73 0.60 Theta (mean) 0.84 - - -
** and * significant at 99% and 95% respectively ( t-student is presented next to correspondent coefficient). a) As all the members of ANDEAN and MERCOSUR belong also to LAIA, we isolate the evolution of trade of the two former RTA in computing the dummies for LAIA as follows (i.e. Soloaga and Winters (2001)) : intra-LAIA=LAIA-ANDEAN-MERCOSUR LAIA imports= LAIA imports-ANDEAN imports-MERCOSUR imports LAIA exports= LAIA exports-ANDEAN exports-MERCOSUR exports.
32
A.8 : Evolution of the RTA dummies estimated in panel and in cross-section over 1962-
1996 (αI, αM and αX).
Evolution of the EFTA dummies estimated in cross-section
-3-2-101234
1962 1966 1970 1974 1978 1982 1986 1990 1994
EFTA-intra EFTA-imports EFTA-exports
Evolution of the EFTA dummies estimated in panel
-3-2-101234
1962 1966 1970 1974 1978 1982 1986 1990 1994
EFTA-intra EFTA-imports EFTA-exports
Evolution of the ASEAN dummies estimated in cross-section
-3-2-101234
1962 1966 1970 1974 1978 1982 1986 1990 1994
ASEAN-intra ASEAN-imports ASEAN-exports
Evolution of the ASEAN dummies estimated in panel
-3-2-101234
1962 1966 1970 1974 1978 1982 1986 1990 1994
ASEAN-intra ASEAN-imports ASEAN-exports Evolution of the LAIA dummies estimated
in cross-section
-3-2-101234
1962 1966 1970 1974 1978 1982 1986 1990 1994
LAIA-intra LAIA-imports LAIA-exports
Evolution of the LAIA dummies estimated in panel
-3-2-101234
1962 1966 1970 1974 1978 1982 1986 1990 1994
LAIA-intra LAIA-imports LAIA-exports
Evolution of the CACM dummies estimated in cross-section
-3-2-101234
1962 1966 1970 1974 1978 1982 1986 1990 1994
CACM-intra CACM-imports CACM-exports
Evolution of the CACM dummies estimated in panel
-3-2-101234
1962 1966 1970 1974 1978 1982 1986 1990 1994
CACM-intra CACM-impo rts CACM-expo rts
Evolution of the ANDEAN dummies estimated in cross-section
-3-2-101234
1962 1966 1970 1974 1978 1982 1986 1990 1994
ANDEAN-intra ANDEAN-impo rts ANDEAN-expo rt
Evolution of the ANDEAN dummies estimated in panel
-3-2-101234
1962 1966 1970 1974 1978 1982 1986 1990 1994
ANDEAN-intra ANDEAN-impo rts ANDEAN-expo
32
References
Aitken, N., “The effect of the EEC and EFTA on European Trade: A Temporal cross-section Analysis,” American Economic Review 63 (December 1973), 881-892.
Anderson, J.E., “A theoretical foundation for the Gravity Equation,” American Economic Review 69 (March 1979), 106-116.
Baier, S.L. and J.H. Bergstrand, “The Growth of World Trade : Tariffs, Transport Costs, and Income Similarity,” Journal of International Economics 53 (February 2001), 1-27.
Bayoumi, T. and B. Eichengreen, “Is Regionalism Simply a Diversion? Evidence from the Evolution of the EC and EFTA,” in Ito, T., Krueger, A., eds., Regionalism versus Multilateral Trade Arrangements, (University of Chicago Press, 1997).
Bergstrand, J.H., “The Gravity Equation in International Trade : Some Microeconomic Foundations and Empirical Evidence,” The Review of Economics and Statistics 67 (August 1985), 474-481.
Bergstrand, J.H., “The Generalized Gravity Equation, Monopolistic Competition, and the Factor-Proportions Theory in International Trade,” The Review of Economics and Statistics 71 (February 1989), 143-153.
Breusch, T., G. Mizon and P. Schmidt, “Efficient Estimation Using Panel Data,” Econometrica 57 (1989), 695-700.
Cheng, I.H. and H.J. Wall, “Controlling for Heterogeneity in Gravity Models of Trade,” The Federal Reserve Bank of St. Louis Working Paper No.99-010 A, 1999.
Deardorff, A., “Determinants of Bilateral Trade : Does Gravity Work in a Neoclassical World ?,” in J.A. Frankel, eds., The Regionalization of the World Economy (University of Chicago Press, 1998).
Endoh, M., “Trade Creation and Trade Diversion in the EEC, the LAFTA and the CMEA: 1960-1994,” Applied Economics 31 (February 1999), 207-216.
Egger, P., “An Econometric View on the estimation of Gravity Models and the Calculation of Trade Potentials,” The World Economy 25 (February 2002), 297-312.
Egger, P. and M. Pfaffermayr, “The Proper Econometric Specification of the Gravity Equation : A Three-Way Model with Bilateral Interaction Effects,” WIFO Working Paper, Vienna, 2000.
Frankel, J., Regional Trading Blocs in the World Economic System, (Institute for International Economics, Washington, DC., 1997).
Guillotin, Y. and P. Sevestre, “Estimations de fonctions de gains sur données de panel: endogéneité du capital humain et effets de la sélection,” Economies et Prévision, 116 (May 1994), 119-135.
Hausman, A. and E. Taylor, “Panel data and unobservable individual effects,” Econometrica 49 (November 1981), 1377-1398.
Krueger, A.O., “Trade Creation and Trade Diversion under NAFTA”, NBER Working Paper No. 7429 (New-York: National Bureau of Economic Research), December 1999.
33
Limao, N. and A.J. Venables, “Infrastructure, Geographical Disadvantage and Transport Costs,” World Bank Economic Review 15, (July 2001), 451-479.
Matyas, L., “Proper econometric specification of the Gravity Model,” The World Economy 20 (May 1997), 363-368.
Nijman, T. and M. Verbeek, “Incomplete Panels and Selection Bias,” in L. Matyas and P. Sevestre, eds., The Econometrics of Panel Data (Kluwer. 1992), 262-302.
Polak, J.J., “Is APEC a Natural Regional Trading Bloc,” The World Economy 19 (September 1996), 533-543.
Soloaga, I. and A. Winters, “How Has Regionalism in the 1990s Affected Trade?,” North American Journal of Economics and Finance 12 (2001), 1-29.
Winters, A., “Regionalism and the rest of the world: the irrelevance of the Kemp-Wan theorem”, Oxford Economic Papers 49, (April 1997) 228-234.
Yeats, A., “Does Mercosur's Trade Performance Raise Concerns about the Effects of Regional Trade Arrangements?,” The World Bank Economic Review 12 (January 1998), 1-28.
33